Fuchsia

Flatland timeline

In a typical frame, the order of events when using the Flatland API goes like this[^1]:

  1. The client calls Flatland::Present().
  2. Scenic waits until PresentArgs::requested_presentation_time has been reached and PresentArgs::server_wait_fences have been signaled.
    • requested_presentation_time is the earliest possible time the present may appear on the display.
    • server_wait_fences is a set of zircon events (“fences”) which must all be signaled before the content submitted with the Present() call may appear on the display.
  3. Scenic composites the content from all clients this frame and sends it off to the display.
  4. Scenic sends Flatland::->OnNextFrameBegin() to the client.
  5. The client starts creating their next frame.
  6. Once the image has actually been displayed Scenic sends Flatland::->OnFramePresented() to the client.
  7. Repeat from 1.

This document explains what each of these signals mean.

[^1]: The exact order of these events is not guaranteed. For example ->OnNextFrameBegin() may happen after ->OnFramePresented(), or the client may call Present() again before receiving ->OnFramePresented(). For most clients these edge cases should not be a concern. They should just use the signal from ->OnNextFrameBegin() to draw their next frame and use ->OnNextFramePresented() to collect data, in which case there should be no conflicts.

Present()

As clients make Flatland API calls, those operations are not actually applied right away. Instead these updates are bundled and committed atomically when Present() is called. The set of updates between two calls to Present() is referred to in this document as a present.

When calling Present() the client passes a table of arguments, PresentArgs, which contain arguments for how the Present() call should be handed. The most important arguments for deciding when the present appears on the display are the requested_presentation_time and server_wait_fences.

For more information on how Present() calls are handled, see frame scheduling.

->OnNextFrameBegin()

->OnNextFrameBegin() is sent as a hint to the client for when they should begin creating their next frame. It does not explicitly say anything about the state of Scenic, except that Scenic thinks resource contention should be relatively low. OnNextFrameBegin() is only sent after the client have made previous Present() calls, and only when the client has present credits remaining.

->OnFramePresented()

->OnFramePresented() is sent when one or more presents have actually appeared on the display (i.e. after Scenic receives the VSync signal). It provides two main pieces of information:

  • Timing feedback to the client so they can evaluate how their frame scheduling strategy is working (latency, dropped frames, etc).
  • It is signal for when the effects of a present has updated the state of the ViewTree). After ->OnFramePresented() Scenic guarantees that any subsequent calls to APIs that interact with the ViewTree (such as those in Flatland::ViewBoundProtocols) will act as if the previous present has been applied (though more presents may have been applied afterwards).